Coarse-grained machine learning potential for mesoscale multilayered graphene

A coarse-grained neuroevolution potential (CGNEP) for multilayered graphene based on an ab initio accuracy dataset is developed for mesoscale molecular dynamics simulations. The information loss in coarsening process is discussed and divided into intralayer part and interlayer part. The CGNEP describes the interlayer shear introduced by van der Waals interactions well by modifying the descriptor of NEP. The mechanical properties and vibration frequencies of structures of different sizes are well predicted via CGNEP. Compared with the traditional empirical CG potential, the CGNEP possesses interlayer properties of the structure of graphene and maintains the ability for higher mapping ratio coarsening. The frequencies of a 12-layer graphene membrane with a length and width of 1 μm are directly calculated via the CGNEP with a 64:1 mapping ratio and compared with the experimental results. The proposed CGNEP may be further used for other multilayered CG 2D materials.